I have an NVIDIA Jetson Xavier NX board, and I have successfully installed NVIDIA SDK components.
Now, I have to run multiple repositories for different tasks such as detection, tracking, segmentation, and so on. Since various repositories require different versions of dependencies such as PyTorch, TensorFlow, OpenCV, etc., sometimes even different Cuda Toolkit versions are required. In this view, how can I proceed with the installation procedure so that there is no conflict between the libraries of the various repositories? For instance, on laptop machines, we create different Conda environments. Similarly, how should we proceed on the Jetson development board? Do I have to create the Docker images for every repository that is to be used, or can I simply install them on Jetson Terminal, or any recommended procedure to start with?
I shall be grateful for guidance at this basic level.